Analysis setting for running large models

Dear Forum,

I try to run a problem which has several hundreds of thousands quadratic elements using the Lanczos solver.
I tried to submit using SMP and 8 cpus. I only asked for 20 first eigenmodes.
After 60 minutes I gave up and aborted the job.
With another solver, using equivalent setup, I received results after less than 10 minutes.
I'm sure that I can have faster results using command line arguments.

Does anybody have some experience with a similar problem?

Thank you,
Amir